Re: What's cooking in git.git (Mar 2021, #03; Wed, 10)

[Junio C Hamano <gitster@pobox.com>]

Junio C Hamano <gitster@pobox.com> writes:

>>> * ab/make-cocci-dedup (2021-03-05) 4 commits
>>>  - Makefile/coccicheck: set SPATCH_BATCH_SIZE to 8
>>>  - Makefile/coccicheck: allow for setting xargs concurrency
>>>  - Makefile/coccicheck: speed up and fix bug with duplicate hunks
>>>  - Makefile/coccicheck: add comment heading for all SPATCH flags
>>>
>>>  An attempt to speed up the coccicheck target with incorrect
>>>  results.
>>>
>>>  A reroll exists to address correctness issue, but not picked up.
>>
>> Any reason for not picked up other than "rc period etc...".
>
> As I always say, please don't read anything more than "I happen to
> have seen it" in being in 'seen'.  And that does not even mean
> everything I saw would be on 'seen'.  Especially during the
> pre-release freeze.  I may have time to pick up a replacement for a
> topic that is already in 'seen', to make sure there aren't unexpected
> new conflicts I'll later have to resolve, and if it is too bad, I may
> even drop the old iteration (because it is stale and a new one exists)
> and the new iteration (because it may be fresher but does not work
> well with others).
>
>> I'm
>> confident the patch at
>> https://lore.kernel.org/git/20210306192525.15197-1-avarab@gmail.com/
>> addresses the intra-series bug, and the whole thing solves outstanding
>> bugs on master.
>
> I recall seeing you use a new option to coccinelle that I did not
> get any hit on my search engine in the updated series.  Is the world
> ready for the thing?

So, I had a chance to go back to the list discussion to double check
the details of what I vaguely recalled when I wrote it.

In [v2 5/4] we revert an earlier change to use "--no-includes" in
[v2 2/4] (which got "This part still doesn't make any sense to me."
by Peff) back to use "--all-includes".  If we were fixing an earlier
regression in a hurry, such a messy history may have to be accepted,
but because we are not in a hurry, I'd rather see the series
straightened up.

We start using "--include-headers-for-types" on SPATCH_FLAGS while
doing so, which I didn't find any hits, but a fresh search found
this http://lira.epac.to:8080/doc/coccinelle-doc/manual.pdf and it
seems that at least at release 1.0.4 (Jan 6, 2019) the option is
available.  I couldn't find a reference for the oldest version that
we can safely use, though.

For now, I'll drop the one listed in the "What's cooking" report you
cited above (because it is stale and a new one is coming), and will
wait for a new iteration in a cleaned-up form (because we do not
want to see "ok, let's do no-includes" followed by "ah, that was a
mistake, let's use all-includes like before").

Thanks.